Side Chats and Conversation Search
02:00 · July 10, 2026 · Cursor Blog

Summary
Cursor 3.11 adds side chats that run in parallel with the primary agent thread. Developers can open a side chat via the /side or /btw commands or the chat-panel plus button; each instance inherits context from the main conversation yet remains a complete, persistent agent session. Side chats default to read-only and search-oriented tasks, letting users request clarifications, compare alternatives, or sanity-check decisions while the main agent continues execution.
Conversation search now indexes full agent transcripts locally. From the Agents Window or the command palette, users can query across thousands of past sessions with sub-second latency. Within an open transcript, Cmd+F supports match navigation and a live counter, removing the need to rely solely on conversation titles or PR references.
Project and repository pickers have been consolidated. Scoped search limits results to the current environment—local machine, cloud workspace, or remote host—while a single menu now handles project creation, GitHub or GitLab connection, and branch selection without leaving the picker. Recent items can be cleared inline, and multi-root workspaces are toggled directly inside the flyout.
New cloud-agent hooks expose the conversation lifecycle itself. Callbacks such as beforeSubmitPrompt, afterAgentThought, afterAgentResponse, subagentStart, and stop allow teams to inspect prompts, reasoning traces, and subagent activity, then intervene programmatically. These primitives support construction of self-correcting loops that observe output, detect issues, and adjust subsequent turns without manual intervention.
Why it matters
This update is highly relevant for AI product teams and builders as it directly improves the developer experience in Cursor, a leading AI IDE. The new cloud agent hooks provide actionable ways to build, observe, and refine complex agentic workflows and self-correcting loops.



